Hybrid Approach to Features Extraction for Online Arabic Character Recognition

Abstract

Recognition of Arabic Character field has been gaining more interest for many years, and a large number of research papers and reports have already been published in this area. There are several major issues with Arabic character recognition: Arabic characters are spelled differently (depending on whether they are isolated, at the beginning, in the middle or at the end of the word), multiple characters can have the same body but a number and/or position of various diacritics. The size of the Arabic characters may vary from one writer to another and even within the writing of a single writer, etc. This paper presents a new approach for feature extraction step of online handwritten Arabic character using global and local features. The system was tested with 2000 Characters written by different writers and the best rate of recognition obtained was 92.43%.

“0” and “O” Recognition Based on Deep Learning

“…The recognition of 10,000 long images takes only 3.86 seconds, with a recognition rate of 99.09%. Compared to the references [5][6][7][8][9], not only the recognition rate is improved greatly, but the recognition duration is also shorter, which can fully meet the online recognition requirements.…”

“…S. Mandal et al [7] put forward a frequent counting-based method utilizing the first, second derivatives and the slope and baseline characteristics, which showed an accuracy of 95.1% for handwritten character recognition. The accuracy rate of a SVM-based approach proposed by H. Nakkach et al [8] was 92.43% for the identification and classification of Arabic characters. The deep learning-based convolutional neural network put forward by Shuye Zhang et al [9] exhibited an accuracy of 98.44% for the recognition classification of similar Chinese characters.…”

Real-Time Pashto Handwritten Character Recognition Using Salient Geometric and Spectral Features

“…Large number of systems use Hidden Markov Model (HMM) for real-time handwriting character recognition [15], [33], [5], [14], [8], [19], [17], [20], [4], [2]. The time complexity of these techniques is less than other state-of-theart techniques due to the probabilistic approach for the transition from one state to the other.…”